1. Field of the Invention
The present invention relates to an apparatus and method for recognizing a gesture, and more particularly, to an apparatus and method for recognizing a gesture in an image processing system such as a TV, a home robot, and a game.
2. Description of the Related Art
Demands for an image processing technology are increasing, which can control devices such as TVs, home robots and games by recognizing a human gesture from an image captured by an image capture device such as a PC (personal computer) camera, a mobile phone camera, and a CCD (charge-coupled device) camera.
For the actual application of the image processing technology to a home environment, the image processing technology for gesture recognition must meet certain conditions. As for the conditions, when recognizing a gesture, the image processing technology must not affected by factors such as lightings, specific backgrounds, and ages, hair styles, makeup and various poses of a user, and must have real-time processing potential.
Recently, various technologies for recognizing a gesture have been developed. Gesture recognition methods of those technologies include a method of recognizing a gesture by continuously obtaining differential images with respect to successive images, a method of using optical flow, and a method of detecting and analyzing a hand area using a color model. Although recognizing a gesture, those methods are problematic in that they satisfy only portions of the above-mentioned conditions, not all of the conditions.
Specifically, the method of obtaining only differential images is problematic in that the method application is difficult in the case where a background is complex or there is another person who is moving. Also, the method of detecting and analyzing an edge of the hand area is disadvantageous when a clothing color of a user is similar to a background color. Also, the method using the optical flow and a method of extracting a body silhouette undesirably consume long time, making the real-time processing difficult. Also, a method of dividing a gesture area using a color model is problematic in that sensitivity to the lightings or backgrounds is too high because the method depends only on colors.
Since a hand area is extracted based only on skin information of the hand area, hand-area detection rates of the conventional methods considerably decrease when the background environment changes or and a user changes a clothing style. Hence, an image processing system employing those conventional methods is not able to clearly recognize a human gesture, thereby experiencing mal-functioning in controlling devices.